Reproduce “The 2023 Video Similarity Dataset and Challenge” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Ed Pizzi, Giorgos Kordopatis-Zilos, Hiral Patel, Gheorghe Postelnicu, Sugosh Nagavara Ravindra, Akshay Gupta, Symeon Papadopoulos, Giorgos Tolias and 1 more

Researchers from Meta AI and collaborating universities introduced the DVSC23 dataset and challenge for video copy detection and localization. This initiative provides a large-scale benchmark and evaluation framework for identifying and temporally segmenting copied content within videos, with top-performing methods achieving VCD 𝜇AP scores up to 87.2% and VCL 𝜇AP scores up to 91.5%.

Key insights

  • Frame-level feature extraction, rather than global video descriptors, is essential for robustly detecting partial video copies.
  • Score normalization, which accounts for background similarity, and temporal network-based aggregation methods are effective strategies for improving both detection and localization accuracy.
  • Specialized classifiers for detecting edited query videos significantly help in filtering distractors and enhancing robustness against multiple transformations.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.